busybox/networking/zcip.c
<<
>>
Prefs
   1/* vi: set sw=4 ts=4: */
   2/*
   3 * RFC3927 ZeroConf IPv4 Link-Local addressing
   4 * (see <http://www.zeroconf.org/>)
   5 *
   6 * Copyright (C) 2003 by Arthur van Hoff (avh@strangeberry.com)
   7 * Copyright (C) 2004 by David Brownell
   8 *
   9 * Licensed under GPLv2 or later, see file LICENSE in this source tree.
  10 */
  11/*
  12 * ZCIP just manages the 169.254.*.* addresses.  That network is not
  13 * routed at the IP level, though various proxies or bridges can
  14 * certainly be used.  Its naming is built over multicast DNS.
  15 */
  16//config:config ZCIP
  17//config:       bool "zcip (8.4 kb)"
  18//config:       default y
  19//config:       select PLATFORM_LINUX
  20//config:       select FEATURE_SYSLOG
  21//config:       help
  22//config:       ZCIP provides ZeroConf IPv4 address selection, according to RFC 3927.
  23//config:       It's a daemon that allocates and defends a dynamically assigned
  24//config:       address on the 169.254/16 network, requiring no system administrator.
  25//config:
  26//config:       See http://www.zeroconf.org for further details, and "zcip.script"
  27//config:       in the busybox examples.
  28
  29//applet:IF_ZCIP(APPLET(zcip, BB_DIR_SBIN, BB_SUID_DROP))
  30
  31//kbuild:lib-$(CONFIG_ZCIP) += zcip.o
  32
  33//#define DEBUG
  34
  35// TODO:
  36// - more real-world usage/testing, especially daemon mode
  37// - kernel packet filters to reduce scheduling noise
  38// - avoid silent script failures, especially under load...
  39// - link status monitoring (restart on link-up; stop on link-down)
  40
  41//usage:#define zcip_trivial_usage
  42//usage:       "[OPTIONS] IFACE SCRIPT"
  43//usage:#define zcip_full_usage "\n\n"
  44//usage:       "Manage a ZeroConf IPv4 link-local address\n"
  45//usage:     "\n        -f              Run in foreground"
  46//usage:     "\n        -q              Quit after obtaining address"
  47//usage:     "\n        -r 169.254.x.x  Request this address first"
  48//usage:     "\n        -l x.x.0.0      Use this range instead of 169.254"
  49//usage:     "\n        -v              Verbose"
  50//usage:     "\n"
  51//usage:     "\n$LOGGING=none           Suppress logging"
  52//usage:     "\n$LOGGING=syslog         Log to syslog"
  53//usage:     "\n"
  54//usage:     "\nWith no -q, runs continuously monitoring for ARP conflicts,"
  55//usage:     "\nexits only on I/O errors (link down etc)"
  56
  57#include "libbb.h"
  58#include "common_bufsiz.h"
  59#include <netinet/ether.h>
  60#include <net/if.h>
  61#include <net/if_arp.h>
  62#include <linux/sockios.h>
  63
  64#include <syslog.h>
  65
  66/* We don't need more than 32 bits of the counter */
  67#define MONOTONIC_US() ((unsigned)monotonic_us())
  68
  69struct arp_packet {
  70        struct ether_header eth;
  71        struct ether_arp arp;
  72} PACKED;
  73
  74enum {
  75        /* 0-1 seconds before sending 1st probe */
  76        PROBE_WAIT = 1,
  77        /* 1-2 seconds between probes */
  78        PROBE_MIN = 1,
  79        PROBE_MAX = 2,
  80        PROBE_NUM = 3,          /* total probes to send */
  81        ANNOUNCE_INTERVAL = 2,  /* 2 seconds between announces */
  82        ANNOUNCE_NUM = 3,       /* announces to send */
  83        /* if probe/announce sees a conflict, multiply RANDOM(NUM_CONFLICT) by... */
  84        CONFLICT_MULTIPLIER = 2,
  85        /* if we monitor and see a conflict, how long is defend state? */
  86        DEFEND_INTERVAL = 10,
  87};
  88
  89/* States during the configuration process. */
  90enum {
  91        PROBE = 0,
  92        ANNOUNCE,
  93        MONITOR,
  94        DEFEND
  95};
  96
  97#define VDBG(...) do { } while (0)
  98
  99
 100enum {
 101        sock_fd = 3
 102};
 103
 104struct globals {
 105        struct sockaddr iface_sockaddr;
 106        struct ether_addr our_ethaddr;
 107        uint32_t localnet_ip;
 108} FIX_ALIASING;
 109#define G (*(struct globals*)bb_common_bufsiz1)
 110#define INIT_G() do { setup_common_bufsiz(); } while (0)
 111
 112
 113/**
 114 * Pick a random link local IP address on 169.254/16, except that
 115 * the first and last 256 addresses are reserved.
 116 */
 117static uint32_t pick_nip(void)
 118{
 119        unsigned tmp;
 120
 121        do {
 122                tmp = rand() & IN_CLASSB_HOST;
 123        } while (tmp > (IN_CLASSB_HOST - 0x0200));
 124        return htonl((G.localnet_ip + 0x0100) + tmp);
 125}
 126
 127static const char *nip_to_a(uint32_t nip)
 128{
 129        struct in_addr in;
 130        in.s_addr = nip;
 131        return inet_ntoa(in);
 132}
 133
 134/**
 135 * Broadcast an ARP packet.
 136 */
 137static void send_arp_request(
 138        /* int op, - always ARPOP_REQUEST */
 139        /* const struct ether_addr *source_eth, - always &G.our_ethaddr */
 140                                        uint32_t source_nip,
 141        const struct ether_addr *target_eth, uint32_t target_nip)
 142{
 143        enum { op = ARPOP_REQUEST };
 144#define source_eth (&G.our_ethaddr)
 145
 146        struct arp_packet p;
 147        memset(&p, 0, sizeof(p));
 148
 149        // ether header
 150        p.eth.ether_type = htons(ETHERTYPE_ARP);
 151        memcpy(p.eth.ether_shost, source_eth, ETH_ALEN);
 152        memset(p.eth.ether_dhost, 0xff, ETH_ALEN);
 153
 154        // arp request
 155        p.arp.arp_hrd = htons(ARPHRD_ETHER);
 156        p.arp.arp_pro = htons(ETHERTYPE_IP);
 157        p.arp.arp_hln = ETH_ALEN;
 158        p.arp.arp_pln = 4;
 159        p.arp.arp_op = htons(op);
 160        memcpy(&p.arp.arp_sha, source_eth, ETH_ALEN);
 161        memcpy(&p.arp.arp_spa, &source_nip, 4);
 162        memcpy(&p.arp.arp_tha, target_eth, ETH_ALEN);
 163        memcpy(&p.arp.arp_tpa, &target_nip, 4);
 164
 165        // send it
 166        // Even though sock_fd is already bound to G.iface_sockaddr, just send()
 167        // won't work, because "socket is not connected"
 168        // (and connect() won't fix that, "operation not supported").
 169        // Thus we sendto() to G.iface_sockaddr. I wonder which sockaddr
 170        // (from bind() or from sendto()?) kernel actually uses
 171        // to determine iface to emit the packet from...
 172        xsendto(sock_fd, &p, sizeof(p), &G.iface_sockaddr, sizeof(G.iface_sockaddr));
 173#undef source_eth
 174}
 175
 176/**
 177 * Run a script.
 178 * argv[0]:intf argv[1]:script_name argv[2]:junk argv[3]:NULL
 179 */
 180static int run(char *argv[3], const char *param, uint32_t nip)
 181{
 182        int status;
 183        const char *addr = addr; /* for gcc */
 184        const char *fmt = "%s %s %s" + 3;
 185        char *env_ip = env_ip;
 186
 187        argv[2] = (char*)param;
 188
 189        VDBG("%s run %s %s\n", argv[0], argv[1], argv[2]);
 190
 191        if (nip != 0) {
 192                addr = nip_to_a(nip);
 193                /* Must not use setenv() repeatedly, it leaks memory. Use putenv() */
 194                env_ip = xasprintf("ip=%s", addr);
 195                putenv(env_ip);
 196                fmt -= 3;
 197        }
 198        bb_info_msg(fmt, argv[2], argv[0], addr);
 199        status = spawn_and_wait(argv + 1);
 200        if (nip != 0)
 201                bb_unsetenv_and_free(env_ip);
 202
 203        if (status < 0) {
 204                bb_perror_msg("%s %s %s" + 3, argv[2], argv[0]);
 205                return -errno;
 206        }
 207        if (status != 0)
 208                bb_error_msg("script %s %s failed, exitcode=%d", argv[1], argv[2], status & 0xff);
 209        return status;
 210}
 211
 212/**
 213 * Return milliseconds of random delay, up to "secs" seconds.
 214 */
 215static ALWAYS_INLINE unsigned random_delay_ms(unsigned secs)
 216{
 217        return (unsigned)rand() % (secs * 1000);
 218}
 219
 220/**
 221 * main program
 222 */
 223int zcip_main(int argc, char **argv) MAIN_EXTERNALLY_VISIBLE;
 224int zcip_main(int argc UNUSED_PARAM, char **argv)
 225{
 226        char *r_opt;
 227        const char *l_opt = "169.254.0.0";
 228        int state;
 229        int nsent;
 230        unsigned opts;
 231
 232        // Ugly trick, but I want these zeroed in one go
 233        struct {
 234                const struct ether_addr null_ethaddr;
 235                struct ifreq ifr;
 236                uint32_t chosen_nip;
 237                int conflicts;
 238                int timeout_ms; // must be signed
 239                int verbose;
 240        } L;
 241#define null_ethaddr (L.null_ethaddr)
 242#define ifr          (L.ifr         )
 243#define chosen_nip   (L.chosen_nip  )
 244#define conflicts    (L.conflicts   )
 245#define timeout_ms   (L.timeout_ms  )
 246#define verbose      (L.verbose     )
 247
 248        memset(&L, 0, sizeof(L));
 249        INIT_G();
 250
 251#define FOREGROUND (opts & 1)
 252#define QUIT       (opts & 2)
 253        // Parse commandline: prog [options] ifname script
 254        // exactly 2 args; -v accumulates and implies -f
 255        opts = getopt32(argv, "^" "fqr:l:v" "\0" "=2:vv:vf",
 256                                &r_opt, &l_opt, &verbose
 257        );
 258#if !BB_MMU
 259        // on NOMMU reexec early (or else we will rerun things twice)
 260        if (!FOREGROUND)
 261                bb_daemonize_or_rexec(0 /*was: DAEMON_CHDIR_ROOT*/, argv);
 262#endif
 263        // Open an ARP socket
 264        // (need to do it before openlog to prevent openlog from taking
 265        // fd 3 (sock_fd==3))
 266        xmove_fd(xsocket(AF_PACKET, SOCK_PACKET, htons(ETH_P_ARP)), sock_fd);
 267        if (!FOREGROUND) {
 268                // do it before all bb_xx_msg calls
 269                openlog(applet_name, 0, LOG_DAEMON);
 270                logmode |= LOGMODE_SYSLOG;
 271        }
 272        bb_logenv_override();
 273
 274        { // -l n.n.n.n
 275                struct in_addr net;
 276                if (inet_aton(l_opt, &net) == 0
 277                 || (net.s_addr & htonl(IN_CLASSB_NET)) != net.s_addr
 278                ) {
 279                        bb_error_msg_and_die("invalid network address");
 280                }
 281                G.localnet_ip = ntohl(net.s_addr);
 282        }
 283        if (opts & 4) { // -r n.n.n.n
 284                struct in_addr ip;
 285                if (inet_aton(r_opt, &ip) == 0
 286                 || (ntohl(ip.s_addr) & IN_CLASSB_NET) != G.localnet_ip
 287                ) {
 288                        bb_error_msg_and_die("invalid link address");
 289                }
 290                chosen_nip = ip.s_addr;
 291        }
 292        argv += optind - 1;
 293
 294        /* Now: argv[0]:junk argv[1]:intf argv[2]:script argv[3]:NULL */
 295        /* We need to make space for script argument: */
 296        argv[0] = argv[1];
 297        argv[1] = argv[2];
 298        /* Now: argv[0]:intf argv[1]:script argv[2]:junk argv[3]:NULL */
 299#define argv_intf (argv[0])
 300
 301        xsetenv("interface", argv_intf);
 302
 303        // Initialize the interface (modprobe, ifup, etc)
 304        if (run(argv, "init", 0))
 305                return EXIT_FAILURE;
 306
 307        // Initialize G.iface_sockaddr
 308        // G.iface_sockaddr is: { u16 sa_family; u8 sa_data[14]; }
 309        //memset(&G.iface_sockaddr, 0, sizeof(G.iface_sockaddr));
 310        //TODO: are we leaving sa_family == 0 (AF_UNSPEC)?!
 311        safe_strncpy(G.iface_sockaddr.sa_data, argv_intf, sizeof(G.iface_sockaddr.sa_data));
 312
 313        // Bind to the interface's ARP socket
 314        xbind(sock_fd, &G.iface_sockaddr, sizeof(G.iface_sockaddr));
 315
 316        // Get the interface's ethernet address
 317        //memset(&ifr, 0, sizeof(ifr));
 318        strncpy_IFNAMSIZ(ifr.ifr_name, argv_intf);
 319        xioctl(sock_fd, SIOCGIFHWADDR, &ifr);
 320        memcpy(&G.our_ethaddr, &ifr.ifr_hwaddr.sa_data, ETH_ALEN);
 321
 322        // Start with some stable ip address, either a function of
 323        // the hardware address or else the last address we used.
 324        // we are taking low-order four bytes, as top-order ones
 325        // aren't random enough.
 326        // NOTE: the sequence of addresses we try changes only
 327        // depending on when we detect conflicts.
 328        {
 329                uint32_t t;
 330                move_from_unaligned32(t, ((char *)&G.our_ethaddr + 2));
 331                srand(t);
 332        }
 333        // FIXME cases to handle:
 334        //  - zcip already running!
 335        //  - link already has local address... just defend/update
 336
 337        // Daemonize now; don't delay system startup
 338        if (!FOREGROUND) {
 339#if BB_MMU
 340                bb_daemonize(0 /*was: DAEMON_CHDIR_ROOT*/);
 341#endif
 342                bb_info_msg("start, interface %s", argv_intf);
 343        }
 344
 345        // Run the dynamic address negotiation protocol,
 346        // restarting after address conflicts:
 347        //  - start with some address we want to try
 348        //  - short random delay
 349        //  - arp probes to see if another host uses it
 350        //    00:04:e2:64:23:c2 > ff:ff:ff:ff:ff:ff arp who-has 169.254.194.171 tell 0.0.0.0
 351        //  - arp announcements that we're claiming it
 352        //    00:04:e2:64:23:c2 > ff:ff:ff:ff:ff:ff arp who-has 169.254.194.171 (00:04:e2:64:23:c2) tell 169.254.194.171
 353        //  - use it
 354        //  - defend it, within limits
 355        // exit if:
 356        // - address is successfully obtained and -q was given:
 357        //   run "<script> config", then exit with exitcode 0
 358        // - poll error (when does this happen?)
 359        // - read error (when does this happen?)
 360        // - sendto error (in send_arp_request()) (when does this happen?)
 361        // - revents & POLLERR (link down). run "<script> deconfig" first
 362        if (chosen_nip == 0) {
 363 new_nip_and_PROBE:
 364                chosen_nip = pick_nip();
 365        }
 366        nsent = 0;
 367        state = PROBE;
 368        while (1) {
 369                struct pollfd fds[1];
 370                unsigned deadline_us = deadline_us;
 371                struct arp_packet p;
 372                int ip_conflict;
 373                int n;
 374
 375                fds[0].fd = sock_fd;
 376                fds[0].events = POLLIN;
 377                fds[0].revents = 0;
 378
 379                // Poll, being ready to adjust current timeout
 380                if (!timeout_ms) {
 381                        timeout_ms = random_delay_ms(PROBE_WAIT);
 382                        // FIXME setsockopt(sock_fd, SO_ATTACH_FILTER, ...) to
 383                        // make the kernel filter out all packets except
 384                        // ones we'd care about.
 385                }
 386                if (timeout_ms >= 0) {
 387                        // Set deadline_us to the point in time when we timeout
 388                        deadline_us = MONOTONIC_US() + timeout_ms * 1000;
 389                }
 390
 391                VDBG("...wait %d %s nsent=%u\n",
 392                                timeout_ms, argv_intf, nsent);
 393
 394                n = safe_poll(fds, 1, timeout_ms);
 395                if (n < 0) {
 396                        //bb_perror_msg("poll"); - done in safe_poll
 397                        return EXIT_FAILURE;
 398                }
 399                if (n == 0) { // timed out?
 400                        VDBG("state:%d\n", state);
 401                        switch (state) {
 402                        case PROBE:
 403                                // No conflicting ARP packets were seen:
 404                                // we can progress through the states
 405                                if (nsent < PROBE_NUM) {
 406                                        nsent++;
 407                                        VDBG("probe/%u %s@%s\n",
 408                                                        nsent, argv_intf, nip_to_a(chosen_nip));
 409                                        timeout_ms = PROBE_MIN * 1000;
 410                                        timeout_ms += random_delay_ms(PROBE_MAX - PROBE_MIN);
 411                                        send_arp_request(0, &null_ethaddr, chosen_nip);
 412                                        continue;
 413                                }
 414                                // Switch to announce state
 415                                nsent = 0;
 416                                state = ANNOUNCE;
 417                                goto send_announce;
 418                        case ANNOUNCE:
 419                                // No conflicting ARP packets were seen:
 420                                // we can progress through the states
 421                                if (nsent < ANNOUNCE_NUM) {
 422 send_announce:
 423                                        nsent++;
 424                                        VDBG("announce/%u %s@%s\n",
 425                                                        nsent, argv_intf, nip_to_a(chosen_nip));
 426                                        timeout_ms = ANNOUNCE_INTERVAL * 1000;
 427                                        send_arp_request(chosen_nip, &G.our_ethaddr, chosen_nip);
 428                                        continue;
 429                                }
 430                                // Switch to monitor state
 431                                // FIXME update filters
 432                                run(argv, "config", chosen_nip);
 433                                // NOTE: all other exit paths should deconfig...
 434                                if (QUIT)
 435                                        return EXIT_SUCCESS;
 436                                // fall through: switch to MONITOR
 437                        default:
 438                        // case DEFEND:
 439                        // case MONITOR: (shouldn't happen, MONITOR timeout is infinite)
 440                                // Defend period ended with no ARP replies - we won
 441                                timeout_ms = -1; // never timeout in monitor state
 442                                state = MONITOR;
 443                                continue;
 444                        }
 445                }
 446
 447                // Packet arrived, or link went down.
 448                // We need to adjust the timeout in case we didn't receive
 449                // a conflicting packet.
 450                if (timeout_ms > 0) {
 451                        unsigned diff = deadline_us - MONOTONIC_US();
 452                        if ((int)(diff) < 0) {
 453                                // Current time is greater than the expected timeout time.
 454                                diff = 0;
 455                        }
 456                        VDBG("adjusting timeout\n");
 457                        timeout_ms = (diff / 1000) | 1; // never 0
 458                }
 459
 460                if ((fds[0].revents & POLLIN) == 0) {
 461                        if (fds[0].revents & POLLERR) {
 462                                // FIXME: links routinely go down;
 463                                // this shouldn't necessarily exit.
 464                                bb_error_msg("iface %s is down", argv_intf);
 465                                if (state >= MONITOR) {
 466                                        // Only if we are in MONITOR or DEFEND
 467                                        run(argv, "deconfig", chosen_nip);
 468                                }
 469                                return EXIT_FAILURE;
 470                        }
 471                        continue;
 472                }
 473
 474                // Read ARP packet
 475                if (safe_read(sock_fd, &p, sizeof(p)) < 0) {
 476                        bb_perror_msg_and_die(bb_msg_read_error);
 477                }
 478
 479                if (p.eth.ether_type != htons(ETHERTYPE_ARP))
 480                        continue;
 481                if (p.arp.arp_op != htons(ARPOP_REQUEST)
 482                 && p.arp.arp_op != htons(ARPOP_REPLY)
 483                ) {
 484                        continue;
 485                }
 486#ifdef DEBUG
 487                {
 488                        struct ether_addr *sha = (struct ether_addr *) p.arp.arp_sha;
 489                        struct ether_addr *tha = (struct ether_addr *) p.arp.arp_tha;
 490                        struct in_addr *spa = (struct in_addr *) p.arp.arp_spa;
 491                        struct in_addr *tpa = (struct in_addr *) p.arp.arp_tpa;
 492                        VDBG("source=%s %s\n", ether_ntoa(sha), inet_ntoa(*spa));
 493                        VDBG("target=%s %s\n", ether_ntoa(tha), inet_ntoa(*tpa));
 494                }
 495#endif
 496                ip_conflict = 0;
 497                if (memcmp(&p.arp.arp_sha, &G.our_ethaddr, ETH_ALEN) != 0) {
 498                        if (memcmp(p.arp.arp_spa, &chosen_nip, 4) == 0) {
 499                                // A probe or reply with source_ip == chosen ip
 500                                ip_conflict = 1;
 501                        }
 502                        if (p.arp.arp_op == htons(ARPOP_REQUEST)
 503                         && memcmp(p.arp.arp_spa, &const_int_0, 4) == 0
 504                         && memcmp(p.arp.arp_tpa, &chosen_nip, 4) == 0
 505                        ) {
 506                                // A probe with source_ip == 0.0.0.0, target_ip == chosen ip:
 507                                // another host trying to claim this ip!
 508                                ip_conflict |= 2;
 509                        }
 510                }
 511                VDBG("state:%d ip_conflict:%d\n", state, ip_conflict);
 512                if (!ip_conflict)
 513                        continue;
 514
 515                // Either src or target IP conflict exists
 516                if (state <= ANNOUNCE) {
 517                        // PROBE or ANNOUNCE
 518                        conflicts++;
 519                        timeout_ms = PROBE_MIN * 1000
 520                                + CONFLICT_MULTIPLIER * random_delay_ms(conflicts);
 521                        goto new_nip_and_PROBE;
 522                }
 523
 524                // MONITOR or DEFEND: only src IP conflict is a problem
 525                if (ip_conflict & 1) {
 526                        if (state == MONITOR) {
 527                                // Src IP conflict, defend with a single ARP probe
 528                                VDBG("monitor conflict - defending\n");
 529                                timeout_ms = DEFEND_INTERVAL * 1000;
 530                                state = DEFEND;
 531                                send_arp_request(chosen_nip, &G.our_ethaddr, chosen_nip);
 532                                continue;
 533                        }
 534                        // state == DEFEND
 535                        // Another src IP conflict, start over
 536                        VDBG("defend conflict - starting over\n");
 537                        run(argv, "deconfig", chosen_nip);
 538                        conflicts = 0;
 539                        timeout_ms = 0;
 540                        goto new_nip_and_PROBE;
 541                }
 542                // Note: if we only have a target IP conflict here (ip_conflict & 2),
 543                // IOW: if we just saw this sort of ARP packet:
 544                //  aa:bb:cc:dd:ee:ff > xx:xx:xx:xx:xx:xx arp who-has <chosen_nip> tell 0.0.0.0
 545                // we expect _kernel_ to respond to that, because <chosen_nip>
 546                // is (expected to be) configured on this iface.
 547        } // while (1)
 548#undef argv_intf
 549}
 550